Linux服务器conda环境中SadTalker的TTS模块模型文件缺失问题咨询

云上老周

2026-7-15

问题：SadTalker中TTS模块报错“ValueError: [!] Model file not found in the output path”

在Linux服务器的conda虚拟环境配置人脸驱动视频生成模型SadTalker的可视化界面时，TTS模块触发如下报错，使用pip install TTS更新版本后问题仍未解决：

File "launcher.py", line 204, in <module>
   start()
File "launcher.py", line 198, in start
   demo = sadtalker_demo()
File "/home/jiahuaw/test/SadTalker/SadTalker/app_sadtalker.py", line 49, in sadtalker_demo
   tts_talker = TTSTalker()
File "/home/jiahuaw/test/SadTalker/SadTalker/src/utils/text2speech.py", line 9, in __init__
   self.tts = TTS(model_name)
File "/home/jiahuaw/test/SadTalker/SadTalker/venv/lib/python3.8/site-packages/TTS/api.py", line 289, in __init__
   self.load_tts_model_by_name(model_name, gpu)
File "/home/jiahuaw/test/SadTalker/SadTalker/venv/lib/python3.8/site-packages/TTS/api.py", line 385, in load_tts_model_by_name
   model_path, config_path, vocoder_path, vocoder_config_path, model_dir = self.download_model_by_name(
File "/home/jiahuaw/test/SadTalker/SadTalker/venv/lib/python3.8/site-packages/TTS/api.py", line 348, in download_model_by_name
   model_path, config_path, model_item = self.manager.download_model(model_name)
File "/home/jiahuaw/test/SadTalker/SadTalker/venv/lib/python3.8/site-packages/TTS/utils/manage.py", line 309, in download_model
   output_model_path, output_config_path = self._find_files(output_path)
File "/home/jiahuaw/test/SadTalker/SadTalker/venv/lib/python3.8/site-packages/TTS/utils/manage.py", line 332, in _find_files
   raise ValueError(" [!] Model file not found in the output path")
 ValueError:  [!] Model file not found in the output path

解决方法

手动指定TTS模型路径
打开SadTalker的text2speech.py文件（路径：/home/jiahuaw/test/SadTalker/SadTalker/src/utils/text2speech.py），找到初始化TTS的代码行self.tts = TTS(model_name)，替换为手动指定本地已下载的模型路径：
```
self.tts = TTS(model_path="/path/to/your/tts/model.pth", config_path="/path/to/your/tts/config.json")
```
提前下载对应TTS模型到服务器本地目录，填入正确路径即可。
清理缓存后重新下载模型
1. 定位TTS模型缓存目录，通常为~/.local/share/tts/或虚拟环境内的lib/python3.8/site-packages/TTS/.models/
2. 删除该目录下对应模型的文件夹
3. 激活conda虚拟环境，运行命令手动下载SadTalker默认使用的TTS模型：
```
tts --model_name tts_models/en/ljspeech/tacotron2-DDC_ph
```
  确保下载过程网络稳定，无中断。
安装指定版本的TTS
新版本可能存在兼容性问题，尝试安装SadTalker文档推荐的TTS版本：
```
pip uninstall -y TTS
pip install TTS==0.10.0
```
安装完成后重新启动SadTalker。
检查目录权限
确保虚拟环境对TTS模型下载目录有读写权限，执行：
```
chmod -R 755 ~/.local/share/tts/
```

内容的提问来源于stack exchange，提问作者Adamfranklin